A global survey of 703 developers, engineering managers and executives published today found 42% are working for organizations that are already using artificial intelligence (AI) in their software development processes, while another 30% are currently experimenting with it.
Conducted by Benchmarkit on behalf of Bito, a provider of AI tools for analyzing code, the survey found nearly three-quarters (72%) are using AI in some way, with 77% who began using AI tools within the last six months. Nearly three-quarters (73%) reported their organization used some form of AI several times daily.
AI is already being used in code writing (90%), code review (53%), quality assurance/testing (43%), debugging (37%) and design and architecture (37%), according to the survey. Primary benefits cited include improved code quality (57%), accelerated learning of the codebase (49%) and developer satisfaction (46%).
Primary challenges cited include accuracy of results (70%), followed by data privacy (53%), limited customization (33%), learning curve of new tools (25%), cost implications (24%) and resistance to new tools (21%).
Among the respondents working for organizations not using AI, 83% planned to start using AI in software development in the next 12 months. Nearly three quarters (73%) expect AI will be adopted by most developers in their organization in the next two years. A third of respondents (33%) anticipate an uptick in developer productivity of greater than 60% in the next 18 months because of AI.
Bito CEO Amar Goel said that while it’s apparent great strides have been made with AI, there will always be a need to keep humans in the loop to ensure results are consistent and reliable. The most critical thing DevOps teams can do in the meantime is to, at the very least, start experimenting with AI to determine where it can be of the most use, he added.
In the meantime, the survey suggests more advanced software development teams are taking advantage of AI sooner than rivals that are not as committed to automating software development processes, noted Goel.
It is still early days as far as the usage of AI with software development processes is concerned, but it’s clear developers are using it to write code. The issue that DevOps teams will encounter is many developers are relying on general-purpose generative AI platforms to write that code. Those platforms were trained using examples of code of varying quality collected from across the web, many of which contain vulnerabilities or are flawed in other ways. DevOps teams need to carefully review that code before injecting it into a build.
In addition, a general-purpose AI platform such as ChatGPT is not going to create the exact same code for every similar request. The code provided by the platform that runs perfectly fine one day might not work at all the next day.
In the longer term, the arrival of large language models (LLMs) trained on a narrow range of code will make the result generated by an AI platform both more accurate and consistent. The issue DevOps teams need to address in the short term is making sure they are not overwhelmed by a massive amount of code of varying quality that is now being written much faster than existing DevOps pipelines are designed to support.